DonorsChoose¶
DonorsChoose.org receives hundreds of thousands of project proposals each year for classroom projects in need of funding. Right now, a large number of volunteers is needed to manually screen each submission before it's approved to be posted on the DonorsChoose.org website.
Next year, DonorsChoose.org expects to receive close to 500,000 project proposals. As a result, there are three main problems they need to solve:
- How to scale current manual processes and resources to screen 500,000 projects so that they can be posted as quickly and as efficiently as possible
- How to increase the consistency of project vetting across different volunteers to improve the experience for teachers
- How to focus volunteer time on the applications that need the most assistance
The goal of the competition is to predict whether or not a DonorsChoose.org project proposal submitted by a teacher will be approved, using the text of project descriptions as well as additional metadata about the project, teacher, and school. DonorsChoose.org can then use this information to identify projects most likely to need further review before approval.
About the DonorsChoose Data Set¶
The train.csv data set provided by DonorsChoose contains the following features:
| Feature | Description |
|---|---|
project_id |
A unique identifier for the proposed project. Example: p036502 |
project_title |
Title of the project. Examples:
|
project_grade_category |
Grade level of students for which the project is targeted. One of the following enumerated values:
|
project_subject_categories |
One or more (comma-separated) subject categories for the project from the following enumerated list of values:
Examples:
|
school_state |
State where school is located (Two-letter U.S. postal code). Example: WY |
project_subject_subcategories |
One or more (comma-separated) subject subcategories for the project. Examples:
|
project_resource_summary |
An explanation of the resources needed for the project. Example:
|
project_essay_1 |
First application essay* |
project_essay_2 |
Second application essay* |
project_essay_3 |
Third application essay* |
project_essay_4 |
Fourth application essay* |
project_submitted_datetime |
Datetime when project application was submitted. Example: 2016-04-28 12:43:56.245 |
teacher_id |
A unique identifier for the teacher of the proposed project. Example: bdf8baa8fedef6bfeec7ae4ff1c15c56 |
teacher_prefix |
Teacher's title. One of the following enumerated values:
|
teacher_number_of_previously_posted_projects |
Number of project applications previously submitted by the same teacher. Example: 2 |
* See the section Notes on the Essay Data for more details about these features.
Additionally, the resources.csv data set provides more data about the resources required for each project. Each line in this file represents a resource required by a project:
| Feature | Description |
|---|---|
id |
A project_id value from the train.csv file. Example: p036502 |
description |
Desciption of the resource. Example: Tenor Saxophone Reeds, Box of 25 |
quantity |
Quantity of the resource required. Example: 3 |
price |
Price of the resource required. Example: 9.95 |
Note: Many projects require multiple resources. The id value corresponds to a project_id in train.csv, so you use it as a key to retrieve all resources needed for a project:
The data set contains the following label (the value you will attempt to predict):
| Label | Description |
|---|---|
project_is_approved |
A binary flag indicating whether DonorsChoose approved the project. A value of 0 indicates the project was not approved, and a value of 1 indicates the project was approved. |
Notes on the Essay Data¶
-
Prior to May 17, 2016, the prompts for the essays were as follows:
- __project_essay_1:__ "Introduce us to your classroom"
- __project_essay_2:__ "Tell us more about your students"
- __project_essay_3:__ "Describe how your students will use the materials you're requesting"
- __project_essay_3:__ "Close by sharing why your project will make a difference"
-
Starting on May 17, 2016, the number of essays was reduced from 4 to 2, and the prompts for the first 2 essays were changed to the following:
- __project_essay_1:__ "Describe your students: What makes your students special? Specific details about their background, your neighborhood, and your school are all helpful."
- __project_essay_2:__ "About your project: How will these materials make a difference in your students' learning and improve their school lives?"
For all projects with project_submitted_datetime of 2016-05-17 and later, the values of project_essay_3 and project_essay_4 will be NaN.
In [2]:
%matplotlib inline
import warnings
warnings.filterwarnings("ignore")
import sqlite3
import pandas as pd
import numpy as np
import nltk
import string
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics import confusion_matrix
from sklearn import metrics
from sklearn.metrics import roc_curve, auc
from nltk.stem.porter import PorterStemmer
import re
# Tutorial about Python regular expressions: https://pymotw.com/2/re/
import string
from nltk.corpus import stopwords
from nltk.stem import PorterStemmer
from nltk.stem.wordnet import WordNetLemmatizer
from gensim.models import Word2Vec
from gensim.models import KeyedVectors
import pickle
from tqdm import tqdm
import os
from chart_studio.plotly import plotly
import plotly.offline as offline
import plotly.graph_objs as go
offline.init_notebook_mode()
from collections import Counter
1.1 Reading Data¶
In [3]:
project_data = pd.read_csv('train_data.csv')
resource_data = pd.read_csv('resources.csv')
In [4]:
print("Number of data points in train data", project_data.shape)
print('-'*50)
print("The attributes of data :", project_data.columns.values)
In [5]:
print("Number of data points in train data", resource_data.shape)
print(resource_data.columns.values)
resource_data.head(2)
Out[5]:
In [6]:
project_data=project_data.sample(n=50000)
project_data.shape
Out[6]:
In [7]:
project_data['project_is_approved'].value_counts()
Out[7]:
In [8]:
# how to replace elements in list python: https://stackoverflow.com/a/2582163/4084039
cols = ['Date' if x=='project_submitted_datetime' else x for x in list(project_data.columns)]
#sort dataframe based on time pandas python: https://stackoverflow.com/a/49702492/4084039
project_data['Date'] = pd.to_datetime(project_data['project_submitted_datetime'])
project_data.drop('project_submitted_datetime', axis=1, inplace=True)
project_data.sort_values(by=['Date'], inplace=True)
# how to reorder columns pandas python: https://stackoverflow.com/a/13148611/4084039
project_data = project_data[cols]
project_data.head(2)
Out[8]:
1.2 preprocessing of project_subject_categories¶
In [9]:
print(project_data['project_subject_categories'].head(5))
In [10]:
catogories = list(project_data['project_subject_categories'].values)
# remove special characters from list of strings python: https://stackoverflow.com/a/47301924/4084039
# https://www.geeksforgeeks.org/removing-stop-words-nltk-python/
# https://stackoverflow.com/questions/23669024/how-to-strip-a-specific-word-from-a-string
# https://stackoverflow.com/questions/8270092/remove-all-whitespace-in-a-string-in-python
cat_list = []
for i in catogories:
temp = ""
# consider we have text like this "Math & Science, Warmth, Care & Hunger"
for j in i.split(','): # it will split it in three parts ["Math & Science", "Warmth", "Care & Hunger"]
if 'The' in j.split(): # this will split each of the catogory based on space "Math & Science"=> "Math","&", "Science"
j=j.replace('The','') # if we have the words "The" we are going to replace it with ''(i.e removing 'The')
j = j.replace(' ','') # we are placeing all the ' '(space) with ''(empty) ex:"Math & Science"=>"Math&Science"
temp+=j.strip()+" " #" abc ".strip() will return "abc", remove the trailing spaces
temp = temp.replace('&','_') # we are replacing the & value into
cat_list.append(temp.strip())
project_data['clean_categories'] = cat_list
project_data.drop(['project_subject_categories'], axis=1, inplace=True)
from collections import Counter
my_counter = Counter()
for word in project_data['clean_categories'].values:
my_counter.update(word.split())
cat_dict = dict(my_counter)
sorted_cat_dict = dict(sorted(cat_dict.items(), key=lambda kv: kv[1]))
In [11]:
print(project_data['clean_categories'].head(5))
1.3 preprocessing of project_subject_subcategories¶
In [12]:
print(project_data['project_subject_subcategories'].head(5))
In [13]:
sub_catogories = list(project_data['project_subject_subcategories'].values)
# remove special characters from list of strings python: https://stackoverflow.com/a/47301924/4084039
# https://www.geeksforgeeks.org/removing-stop-words-nltk-python/
# https://stackoverflow.com/questions/23669024/how-to-strip-a-specific-word-from-a-string
# https://stackoverflow.com/questions/8270092/remove-all-whitespace-in-a-string-in-python
sub_cat_list = []
for i in sub_catogories:
temp = ""
# consider we have text like this "Math & Science, Warmth, Care & Hunger"
for j in i.split(','): # it will split it in three parts ["Math & Science", "Warmth", "Care & Hunger"]
if 'The' in j.split(): # this will split each of the catogory based on space "Math & Science"=> "Math","&", "Science"
j=j.replace('The','') # if we have the words "The" we are going to replace it with ''(i.e removing 'The')
j = j.replace(' ','') # we are placeing all the ' '(space) with ''(empty) ex:"Math & Science"=>"Math&Science"
temp +=j.strip()+" "#" abc ".strip() will return "abc", remove the trailing spaces
temp = temp.replace('&','_')
sub_cat_list.append(temp.strip())
project_data['clean_subcategories'] = sub_cat_list
project_data.drop(['project_subject_subcategories'], axis=1, inplace=True)
# count of all the words in corpus python: https://stackoverflow.com/a/22898595/4084039
my_counter = Counter()
for word in project_data['clean_subcategories'].values:
my_counter.update(word.split())
sub_cat_dict = dict(my_counter)
sorted_sub_cat_dict = dict(sorted(sub_cat_dict.items(), key=lambda kv: kv[1]))
In [14]:
print(project_data['clean_subcategories'].head(5))
1.4 preprocessing of school_state¶
In [15]:
my_counter = Counter()
for word in project_data['school_state'].values:
my_counter.update(word.split())
state_dict = dict(my_counter)
sorted_state_dict = dict(sorted(state_dict.items(), key=lambda kv: kv[1]))
In [16]:
print(project_data['school_state'].head(5))
1.5 preprocessing of project_grade_category¶
In [17]:
project_data['project_grade_category'].value_counts()
Out[17]:
In [18]:
#https://stackoverflow.com/questions/36383821/pandas-dataframe-apply-function-to-column-strings-based-on-other-column-value
project_data['project_grade_category'] = project_data['project_grade_category'].str.replace(' ','_')
project_data['project_grade_category'] = project_data['project_grade_category'].str.replace('-','_')
project_data['project_grade_category'] = project_data['project_grade_category'].str.lower()
project_data['project_grade_category'].value_counts()
Out[18]:
In [19]:
my_counter = Counter()
for word in project_data['project_grade_category'].values:
my_counter.update(word.split())
grade_dict = dict(my_counter)
sorted_grade_dict = dict(sorted(grade_dict.items(), key=lambda kv: kv[1]))
In [20]:
print(sorted_grade_dict)
1.6 preprocessing of teacher_prefix¶
In [21]:
project_data['teacher_prefix'].value_counts()
Out[21]:
In [22]:
project_data['teacher_prefix']=project_data['teacher_prefix'].fillna('Mrs.')
project_data['teacher_prefix'] = project_data['teacher_prefix'].str.replace('-','_')
project_data['teacher_prefix'] = project_data['teacher_prefix'].str.replace('.',' ')
project_data['teacher_prefix'] = project_data['teacher_prefix'].str.lower()
project_data['teacher_prefix'].value_counts()
Out[22]:
In [23]:
my_counter = Counter()
for word in project_data['teacher_prefix'].values:
my_counter.update(word.split())
teacher_dict = dict(my_counter)
sorted_teacher_dict = dict(sorted(teacher_dict.items(), key=lambda kv: kv[1]))
In [24]:
print(project_data['teacher_prefix'].head(5))
1.7 Text preprocessing¶
In [25]:
# merge two column text dataframe:
project_data["essay"] = project_data["project_essay_1"].map(str) +\
project_data["project_essay_2"].map(str) + \
project_data["project_essay_3"].map(str) + \
project_data["project_essay_4"].map(str)
In [26]:
project_data.head(2)
Out[26]:
In [27]:
#### 1.4.2.3 Using Pretrained Models: TFIDF weighted W2V
In [28]:
# printing some random reviews
print(project_data['essay'].values[0])
print("="*50)
print(project_data['essay'].values[150])
print("="*50)
print(project_data['essay'].values[1000])
print("="*50)
print(project_data['essay'].values[20000])
print("="*50)
print(project_data['essay'].values[49999])
print("="*50)
In [29]:
# https://stackoverflow.com/a/47091490/4084039
import re
def decontracted(phrase):
# specific
phrase = re.sub(r"won't", "will not", phrase)
phrase = re.sub(r"can\'t", "can not", phrase)
# general
phrase = re.sub(r"n\'t", " not", phrase)
phrase = re.sub(r"\'re", " are", phrase)
phrase = re.sub(r"\'s", " is", phrase)
phrase = re.sub(r"\'d", " would", phrase)
phrase = re.sub(r"\'ll", " will", phrase)
phrase = re.sub(r"\'t", " not", phrase)
phrase = re.sub(r"\'ve", " have", phrase)
phrase = re.sub(r"\'m", " am", phrase)
return phrase
In [30]:
sent = decontracted(project_data['essay'].values[20000])
print(sent)
print("="*50)
In [31]:
# \r \n \t remove from string python: http://texthandler.com/info/remove-line-breaks-python/
sent = sent.replace('\\r', ' ')
sent = sent.replace('\\"', ' ')
sent = sent.replace('\\n', ' ')
print(sent)
In [32]:
#remove spacial character: https://stackoverflow.com/a/5843547/4084039
sent = re.sub('[^A-Za-z0-9]+', ' ', sent)
print(sent)
In [33]:
# https://gist.github.com/sebleier/554280
# we are removing the words from the stop words list: 'no', 'nor', 'not'
stopwords= ['i', 'me', 'my', 'myself', 'we', 'our', 'ours', 'ourselves', 'you', "you're", "you've",\
"you'll", "you'd", 'your', 'yours', 'yourself', 'yourselves', 'he', 'him', 'his', 'himself', \
'she', "she's", 'her', 'hers', 'herself', 'it', "it's", 'its', 'itself', 'they', 'them', 'their',\
'theirs', 'themselves', 'what', 'which', 'who', 'whom', 'this', 'that', "that'll", 'these', 'those', \
'am', 'is', 'are', 'was', 'were', 'be', 'been', 'being', 'have', 'has', 'had', 'having', 'do', 'does', \
'did', 'doing', 'a', 'an', 'the', 'and', 'but', 'if', 'or', 'because', 'as', 'until', 'while', 'of', \
'at', 'by', 'for', 'with', 'about', 'against', 'between', 'into', 'through', 'during', 'before', 'after',\
'above', 'below', 'to', 'from', 'up', 'down', 'in', 'out', 'on', 'off', 'over', 'under', 'again', 'further',\
'then', 'once', 'here', 'there', 'when', 'where', 'why', 'how', 'all', 'any', 'both', 'each', 'few', 'more',\
'most', 'other', 'some', 'such', 'only', 'own', 'same', 'so', 'than', 'too', 'very', \
's', 't', 'can', 'will', 'just', 'don', "don't", 'should', "should've", 'now', 'd', 'll', 'm', 'o', 're', \
've', 'y', 'ain', 'aren', "aren't", 'couldn', "couldn't", 'didn', "didn't", 'doesn', "doesn't", 'hadn',\
"hadn't", 'hasn', "hasn't", 'haven', "haven't", 'isn', "isn't", 'ma', 'mightn', "mightn't", 'mustn',\
"mustn't", 'needn', "needn't", 'shan', "shan't", 'shouldn', "shouldn't", 'wasn', "wasn't", 'weren', "weren't", \
'won', "won't", 'wouldn', "wouldn't"]
In [34]:
# Combining all the above stundents
from tqdm import tqdm
preprocessed_essays = []
# tqdm is for printing the status bar
for sentance in tqdm(project_data['essay'].values):
sent = decontracted(sentance)
sent = sent.replace('\\r', ' ')
sent = sent.replace('\\"', ' ')
sent = sent.replace('\\n', ' ')
sent = re.sub('[^A-Za-z0-9]+', ' ', sent)
# https://gist.github.com/sebleier/554280
sent = ' '.join(e for e in sent.split() if e not in stopwords)
preprocessed_essays.append(sent.lower().strip())
In [35]:
# after preprocesing
preprocessed_essays[20000]
Out[35]:
1.8 Preprocessing of `project_title`
In [36]:
# similarly you can preprocess the titles also
preprocessed_titles = []
# tqdm is for printing the status bar
for sentence in tqdm(project_data['project_title'].values):
sent = decontracted(sentence)
sent = sent.replace('\\r', ' ')
sent = sent.replace('\\"', ' ')
sent = sent.replace('\\n', ' ')
sent = re.sub('[^A-Za-z0-9]+', ' ', sent)
# https://gist.github.com/sebleier/554280
sent = ' '.join(e for e in sent.split() if e not in stopwords)
preprocessed_titles.append(sent.lower().strip())
In [37]:
print(45660,preprocessed_titles[45660])
In [38]:
project_data['project_title']=preprocessed_titles
In [39]:
print(45660,project_data['project_title'].iloc[45660])
Sentiment Analysis of essays
In [40]:
import nltk
nltk.downloader.download('vader_lexicon')
from nltk.sentiment.vader import SentimentIntensityAnalyzer
analyser = SentimentIntensityAnalyzer()
neg = []
pos = []
neu = []
compound = []
for a in tqdm(project_data["essay"]) :
b = analyser.polarity_scores(a)['neg']
c = analyser.polarity_scores(a)['pos']
d = analyser.polarity_scores(a)['neu']
e = analyser.polarity_scores(a)['compound']
neg.append(b)
pos.append(c)
neu.append(d)
compound.append(e)
In [41]:
project_data["pos"] = pos
In [42]:
project_data["neg"] = neg
In [43]:
project_data["neu"] = neu
In [44]:
project_data["compound"] = compound
Number of Words in Title¶
In [45]:
title_word_count = []
for a in project_data["project_title"] :
b = len(a.split())
title_word_count.append(b)
project_data["title_word_count"] = title_word_count
In [46]:
print(title_word_count)
Number of Words in Essays¶
In [47]:
essay_word_count = []
for a in project_data["essay"] :
b = len(a.split())
essay_word_count.append(b)
project_data["essay_word_count"] = essay_word_count
In [48]:
print(essay_word_count)
1.9 Preparing data for models¶
In [49]:
project_data.columns
Out[49]:
we are going to consider
- school_state : categorical data
- clean_categories : categorical data
- clean_subcategories : categorical data
- project_grade_category : categorical data
- teacher_prefix : categorical data
- project_title : text data
- text : text data
- project_resource_summary: text data (optinal)
- quantity : numerical (optinal)
- teacher_number_of_previously_posted_projects : numerical
- price : numericalAssignment 5: Logistic Regression¶
- [Task-1] Logistic Regression(either SGDClassifier with log loss, or LogisticRegression) on these feature sets
- Set 1: categorical, numerical features + project_title(BOW) + preprocessed_eassay (`BOW with bi-grams` with `min_df=10` and `max_features=5000`)
- Set 2: categorical, numerical features + project_title(TFIDF)+ preprocessed_eassay (`TFIDF with bi-grams` with `min_df=10` and `max_features=5000`)
- Set 3: categorical, numerical features + project_title(AVG W2V)+ preprocessed_eassay (AVG W2V)
- Set 4: categorical, numerical features + project_title(TFIDF W2V)+ preprocessed_essay (TFIDF W2V)
- Hyper paramter tuning (find best hyper parameters corresponding the algorithm that you choose)
- Find the best hyper parameter which will give the maximum AUC value
- Find the best hyper paramter using k-fold cross validation or simple cross validation data
- Use gridsearch cv or randomsearch cv or you can also write your own for loops to do this task of hyperparameter tuning
- Representation of results
- You need to plot the performance of model both on train data and cross validation data for each hyper parameter, like shown in the figure.
- Once after you found the best hyper parameter, you need to train your model with it, and find the AUC on test data and plot the ROC curve on both train and test.
- Along with plotting ROC curve, you need to print the confusion matrix with predicted and original labels of test data points. Please visualize your confusion matrices using seaborn heatmaps.
- You need to plot the performance of model both on train data and cross validation data for each hyper parameter, like shown in the figure.
- [Task-2] Apply Logistic Regression on the below feature set Set 5 by finding the best hyper parameter as suggested in step 2 and step 3.
- Consider these set of features Set 5 :
- school_state : categorical data
- clean_categories : categorical data
- clean_subcategories : categorical data
- project_grade_category :categorical data
- teacher_prefix : categorical data
- quantity : numerical data
- teacher_number_of_previously_posted_projects : numerical data
- price : numerical data
- sentiment score's of each of the essay : numerical data
- number of words in the title : numerical data
- number of words in the combine essays : numerical data
- Conclusion
- You need to summarize the results at the end of the notebook, summarize it in the table format. To print out a table please refer to this prettytable library link
- You need to summarize the results at the end of the notebook, summarize it in the table format. To print out a table please refer to this prettytable library link
Note: Data Leakage
- There will be an issue of data-leakage if you vectorize the entire data and then split it into train/cv/test.
- To avoid the issue of data-leakage, make sure to split your data first and then vectorize it.
- While vectorizing your data, apply the method fit_transform() on you train data, and apply the method transform() on cv/test data.
- For more details please go through this link.
2. Logistic Regression
2.1 Splitting data into Train and cross validation(or test): Stratified Sampling
In [50]:
y = project_data['project_is_approved']
print(y.shape)
In [51]:
project_data.drop(['project_is_approved'],axis=1,inplace=True)
In [52]:
X=project_data
print(X.shape)
In [160]:
#train test split
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, stratify=y)
X_train, X_cv, y_train, y_cv = train_test_split(X_train, y_train, test_size=0.33, stratify=y_train)
2.2 Make Data Model Ready: encoding numerical, categorical features
In [161]:
print(X_train.shape, y_train.shape)
print(X_cv.shape, y_cv.shape)
print(X_test.shape, y_test.shape)
print("="*100)
Encoding of Text Data
In [162]:
from sklearn.feature_extraction.text import CountVectorizer
BOW of Essay
In [163]:
vectorizer = CountVectorizer(min_df=10,ngram_range=(2,2), max_features=5000)
In [164]:
vectorizer.fit(X_train['essay'].values) # fit has to happen only on train data
Out[164]:
In [165]:
# we use the fitted CountVectorizer to convert the text to vector
X_train_essay_bow = vectorizer.transform(X_train['essay'].values)
In [166]:
X_cv_essay_bow = vectorizer.transform(X_cv['essay'].values)
In [167]:
X_test_essay_bow = vectorizer.transform(X_test['essay'].values)
In [168]:
print("After vectorizations")
print(X_train_essay_bow.shape, y_train.shape)
print(X_cv_essay_bow.shape, y_cv.shape)
print(X_test_essay_bow.shape, y_test.shape)
print("="*100)
BOW of Title
In [169]:
vectorizer = CountVectorizer(min_df=10,ngram_range=(1,4), max_features=5000)
In [170]:
vectorizer.fit(X_train['project_title'].values) # fit has to happen only on train data
Out[170]:
In [171]:
# we use the fitted CountVectorizer to convert the text to vector
X_train_title_bow = vectorizer.transform(X_train['project_title'].values)
In [172]:
X_cv_title_bow = vectorizer.transform(X_cv['project_title'].values)
In [173]:
X_test_title_bow = vectorizer.transform(X_test['project_title'].values)
In [174]:
print("After vectorizations")
print(X_train_title_bow.shape, y_train.shape)
print(X_cv_title_bow.shape, y_cv.shape)
print(X_test_title_bow.shape, y_test.shape)
print("="*100)
TFIDF of Essay
In [175]:
vectorizer = TfidfVectorizer(min_df=10,ngram_range=(2,2), max_features=5000)
In [176]:
vectorizer.fit(X_train['essay'].values) # fit has to happen only on train data
Out[176]:
In [177]:
# we use the fitted CountVectorizer to convert the text to vector
X_train_essay_tfidf = vectorizer.transform(X_train['essay'].values)
In [178]:
X_cv_essay_tfidf = vectorizer.transform(X_cv['essay'].values)
In [179]:
X_test_essay_tfidf = vectorizer.transform(X_test['essay'].values)
In [180]:
print("After vectorizations")
print(X_train_essay_tfidf.shape, y_train.shape)
print(X_cv_essay_tfidf.shape, y_cv.shape)
print(X_test_essay_tfidf.shape, y_test.shape)
print("="*100)
TFIDF of Title
In [181]:
vectorizer = TfidfVectorizer(min_df=10,ngram_range=(1,4), max_features=5000)
In [182]:
vectorizer.fit(X_train['project_title'].values) # fit has to happen only on train data
Out[182]:
In [183]:
# we use the fitted CountVectorizer to convert the text to vector
X_train_title_tfidf = vectorizer.transform(X_train['project_title'].values)
In [184]:
X_cv_title_tfidf = vectorizer.transform(X_cv['project_title'].values)
In [185]:
X_test_title_tfidf = vectorizer.transform(X_test['project_title'].values)
In [186]:
print("After vectorizations")
print(X_train_title_tfidf.shape, y_train.shape)
print(X_cv_title_tfidf.shape, y_cv.shape)
print(X_test_title_tfidf.shape, y_test.shape)
print("="*100)
Avg W2V of Essay
In [187]:
# stronging variables into pickle files python: http://www.jessicayung.com/how-to-use-pickle-to-save-and-load-variables-in-python/
# make sure you have the glove_vectors file
with open('glove_vectors', 'rb') as f:
model = pickle.load(f,encoding = "ISO-8859-1")
glove_words = set(model.keys())
In [188]:
# average Word2Vec
# compute average word2vec for each essay.
avg_w2v_essay_train = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_train['essay'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
cnt_words =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if word in glove_words:
vector += model[word]
cnt_words += 1
if cnt_words != 0:
vector /= cnt_words
avg_w2v_essay_train.append(vector)
print(len(avg_w2v_essay_train))
print(len(avg_w2v_essay_train[0]))
print(type(avg_w2v_essay_train))
In [189]:
# average Word2Vec
# compute average word2vec for each essay.
avg_w2v_essay_cv = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_cv['essay'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
cnt_words =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if word in glove_words:
vector += model[word]
cnt_words += 1
if cnt_words != 0:
vector /= cnt_words
avg_w2v_essay_cv.append(vector)
print(len(avg_w2v_essay_cv))
print(len(avg_w2v_essay_cv[0]))
print(type(avg_w2v_essay_cv))
In [190]:
# average Word2Vec
# compute average word2vec for each essay.
avg_w2v_essay_test = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_test['essay'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
cnt_words =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if word in glove_words:
vector += model[word]
cnt_words += 1
if cnt_words != 0:
vector /= cnt_words
avg_w2v_essay_test.append(vector)
print(len(avg_w2v_essay_test))
print(len(avg_w2v_essay_test[0]))
print(type(avg_w2v_essay_test))
Avg W2V of Title
In [191]:
# average Word2Vec
# compute average word2vec for each essay.
avg_w2v_title_train = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_train['project_title'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
cnt_words =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if word in glove_words:
vector += model[word]
cnt_words += 1
if cnt_words != 0:
vector /= cnt_words
avg_w2v_title_train.append(vector)
print(len(avg_w2v_title_train))
print(len(avg_w2v_title_train[0]))
print(type(avg_w2v_title_train))
In [192]:
# average Word2Vec
# compute average word2vec for each essay.
avg_w2v_title_cv = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_cv['project_title'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
cnt_words =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if word in glove_words:
vector += model[word]
cnt_words += 1
if cnt_words != 0:
vector /= cnt_words
avg_w2v_title_cv.append(vector)
print(len(avg_w2v_title_cv))
print(len(avg_w2v_title_cv[0]))
print(type(avg_w2v_title_cv))
In [193]:
# average Word2Vec
# compute average word2vec for each essay.
avg_w2v_title_test = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_test['project_title'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
cnt_words =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if word in glove_words:
vector += model[word]
cnt_words += 1
if cnt_words != 0:
vector /= cnt_words
avg_w2v_title_test.append(vector)
print(len(avg_w2v_title_test))
print(len(avg_w2v_title_test[0]))
print(type(avg_w2v_title_test))
TFIDF-W2V of Essay
In [194]:
tfidf_model = TfidfVectorizer(min_df=10,ngram_range=(1,4), max_features=5000)
tfidf_model.fit(X_train['essay'].values)
# we are converting a dictionary with word as a key, and the idf as a value
dictionary = dict(zip(tfidf_model.get_feature_names(), list(tfidf_model.idf_)))
tfidf_words = set(tfidf_model.get_feature_names())
In [195]:
# average Word2Vec
# compute average word2vec for each review.
tfidf_w2v_train_essay = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_train['essay'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
tf_idf_weight =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if (word in glove_words) and (word in tfidf_words):
vec = model[word] # getting the vector for each word
# here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
vector += (vec * tf_idf) # calculating tfidf weighted w2v
tf_idf_weight += tf_idf
if tf_idf_weight != 0:
vector /= tf_idf_weight
tfidf_w2v_train_essay.append(vector)
print(len(tfidf_w2v_train_essay))
print(len(tfidf_w2v_train_essay[0]))
In [196]:
# average Word2Vec
# compute average word2vec for each review.
tfidf_w2v_cv_essay = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_cv['essay'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
tf_idf_weight =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if (word in glove_words) and (word in tfidf_words):
vec = model[word] # getting the vector for each word
# here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
vector += (vec * tf_idf) # calculating tfidf weighted w2v
tf_idf_weight += tf_idf
if tf_idf_weight != 0:
vector /= tf_idf_weight
tfidf_w2v_cv_essay.append(vector)
print(len(tfidf_w2v_cv_essay))
print(len(tfidf_w2v_cv_essay[0]))
In [197]:
# average Word2Vec
# compute average word2vec for each review.
tfidf_w2v_test_essay = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_test['essay'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
tf_idf_weight =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if (word in glove_words) and (word in tfidf_words):
vec = model[word] # getting the vector for each word
# here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
vector += (vec * tf_idf) # calculating tfidf weighted w2v
tf_idf_weight += tf_idf
if tf_idf_weight != 0:
vector /= tf_idf_weight
tfidf_w2v_test_essay.append(vector)
print(len(tfidf_w2v_test_essay))
print(len(tfidf_w2v_test_essay[0]))
TFIDF-W2V of Title
In [198]:
tfidf_model = TfidfVectorizer(min_df=10,ngram_range=(1,4), max_features=5000)
tfidf_model.fit(X_train['project_title'].values)
# we are converting a dictionary with word as a key, and the idf as a value
dictionary = dict(zip(tfidf_model.get_feature_names(), list(tfidf_model.idf_)))
tfidf_words = set(tfidf_model.get_feature_names())
In [199]:
# average Word2Vec
# compute average word2vec for each review.
tfidf_w2v_train_title = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_train['project_title'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
tf_idf_weight =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if (word in glove_words) and (word in tfidf_words):
vec = model[word] # getting the vector for each word
# here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
vector += (vec * tf_idf) # calculating tfidf weighted w2v
tf_idf_weight += tf_idf
if tf_idf_weight != 0:
vector /= tf_idf_weight
tfidf_w2v_train_title.append(vector)
print(len(tfidf_w2v_train_title))
print(len(tfidf_w2v_train_title[0]))
In [200]:
# average Word2Vec
# compute average word2vec for each review.
tfidf_w2v_cv_title = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_cv['project_title'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
tf_idf_weight =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if (word in glove_words) and (word in tfidf_words):
vec = model[word] # getting the vector for each word
# here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
vector += (vec * tf_idf) # calculating tfidf weighted w2v
tf_idf_weight += tf_idf
if tf_idf_weight != 0:
vector /= tf_idf_weight
tfidf_w2v_cv_title.append(vector)
print(len(tfidf_w2v_cv_title))
print(len(tfidf_w2v_cv_title[0]))
In [201]:
# average Word2Vec
# compute average word2vec for each review.
tfidf_w2v_test_title = []; # the avg-w2v for each sentence/review is stored in this list
for sentence in tqdm(X_test['project_title'].values): # for each review/sentence
vector = np.zeros(300) # as word vectors are of zero length
tf_idf_weight =0; # num of words with a valid vector in the sentence/review
for word in sentence.split(): # for each word in a review/sentence
if (word in glove_words) and (word in tfidf_words):
vec = model[word] # getting the vector for each word
# here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))
tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word
vector += (vec * tf_idf) # calculating tfidf weighted w2v
tf_idf_weight += tf_idf
if tf_idf_weight != 0:
vector /= tf_idf_weight
tfidf_w2v_test_title.append(vector)
print(len(tfidf_w2v_test_title))
print(len(tfidf_w2v_test_title[0]))
2.3 Make Data Model Ready: encoding eassay, and project_title
In [202]:
price_data = resource_data.groupby('id').agg({'price':'sum', 'quantity':'sum'}).reset_index()
In [203]:
price_data.head(5)
Out[203]:
In [204]:
X_train=pd.merge(X_train,price_data,on='id',how='left')
X_test=pd.merge(X_test,price_data,on='id',how='left')
X_cv=pd.merge(X_cv,price_data,on='id',how='left')
In [205]:
X_train=X_train.fillna(0)
X_cv=X_cv.fillna(0)
X_test=X_test.fillna(0)
Normalizing the numerical features: Price
In [206]:
from sklearn.preprocessing import Normalizer
normalizer = Normalizer()
# normalizer.fit(X_train['price'].values)
# this will rise an error Expected 2D array, got 1D array instead:
# array=[105.22 215.96 96.01 ... 368.98 80.53 709.67].
# Reshape your data either using
# array.reshape(-1, 1) if your data has a single feature
# array.reshape(1, -1) if it contains a single sample.
normalizer.fit(X_train['price'].values.reshape(-1,1))
X_train_price_norm = normalizer.transform(X_train['price'].values.reshape(-1,1))
X_cv_price_norm = normalizer.transform(X_cv['price'].values.reshape(-1,1))
X_test_price_norm = normalizer.transform(X_test['price'].values.reshape(-1,1))
print("After vectorizations")
print(X_train_price_norm.shape, y_train.shape)
print(X_cv_price_norm.shape, y_cv.shape)
print(X_test_price_norm.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Number of previously posted projects
In [207]:
normalizer = Normalizer()
normalizer.fit(X_train['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))
X_train_project_norm = normalizer.transform(X_train['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))
X_cv_project_norm = normalizer.transform(X_cv['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))
X_test_project_norm = normalizer.transform(X_test['teacher_number_of_previously_posted_projects'].values.reshape(-1,1))
print("After vectorizations")
print(X_train_project_norm.shape, y_train.shape)
print(X_cv_project_norm.shape, y_cv.shape)
print(X_test_project_norm.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Title word Count
In [208]:
normalizer = Normalizer()
normalizer.fit(X_train['title_word_count'].values.reshape(-1,1))
X_train_title_norm = normalizer.transform(X_train['title_word_count'].values.reshape(-1,1))
X_cv_title_norm = normalizer.transform(X_cv['title_word_count'].values.reshape(-1,1))
X_test_title_norm = normalizer.transform(X_test['title_word_count'].values.reshape(-1,1))
print("After vectorizations")
print(X_train_title_norm.shape, y_train.shape)
print(X_cv_title_norm.shape, y_cv.shape)
print(X_test_title_norm.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Essay word Count
In [209]:
normalizer = Normalizer()
normalizer.fit(X_train['essay_word_count'].values.reshape(-1,1))
X_train_essay_norm = normalizer.transform(X_train['essay_word_count'].values.reshape(-1,1))
X_cv_essay_norm = normalizer.transform(X_cv['essay_word_count'].values.reshape(-1,1))
X_test_essay_norm = normalizer.transform(X_test['essay_word_count'].values.reshape(-1,1))
print("After vectorizations")
print(X_train_essay_norm.shape, y_train.shape)
print(X_cv_essay_norm.shape, y_cv.shape)
print(X_test_essay_norm.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Essay Sentiments-Positive
In [210]:
normalizer = Normalizer()
normalizer.fit(X_train['pos'].values.reshape(-1,1))
essay_sent_pos_train = normalizer.transform(X_train['pos'].values.reshape(-1,1))
essay_sent_pos_cv = normalizer.transform(X_cv['pos'].values.reshape(-1,1))
essay_sent_pos_test = normalizer.transform(X_test['pos'].values.reshape(-1,1))
print("After vectorizations")
print(essay_sent_pos_train.shape, y_train.shape)
print(essay_sent_pos_cv.shape, y_cv.shape)
print(essay_sent_pos_test.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Essay Sentiments-Negative
In [211]:
normalizer = Normalizer()
normalizer.fit(X_train['neg'].values.reshape(-1,1))
essay_sent_neg_train = normalizer.transform(X_train['neg'].values.reshape(-1,1))
essay_sent_neg_cv = normalizer.transform(X_cv['neg'].values.reshape(-1,1))
essay_sent_neg_test = normalizer.transform(X_test['neg'].values.reshape(-1,1))
print("After vectorizations")
print(essay_sent_neg_train.shape, y_train.shape)
print(essay_sent_neg_cv.shape, y_cv.shape)
print(essay_sent_neg_test.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Essay Sentiments-Neutral
In [212]:
normalizer = Normalizer()
normalizer.fit(X_train['neu'].values.reshape(-1,1))
essay_sent_neu_train = normalizer.transform(X_train['neu'].values.reshape(-1,1))
essay_sent_neu_cv = normalizer.transform(X_cv['neu'].values.reshape(-1,1))
essay_sent_neu_test = normalizer.transform(X_test['neu'].values.reshape(-1,1))
print("After vectorizations")
print(essay_sent_neu_train.shape, y_train.shape)
print(essay_sent_neu_cv.shape, y_cv.shape)
print(essay_sent_neu_test.shape, y_test.shape)
print("="*100)
Normalizing the numerical features: Essay Sentiments-Compound
In [213]:
normalizer = Normalizer()
normalizer.fit(X_train['compound'].values.reshape(-1,1))
essay_sent_comp_train = normalizer.transform(X_train['compound'].values.reshape(-1,1))
essay_sent_comp_cv = normalizer.transform(X_cv['compound'].values.reshape(-1,1))
essay_sent_comp_test = normalizer.transform(X_test['compound'].values.reshape(-1,1))
print("After vectorizations")
print(essay_sent_comp_train.shape, y_train.shape)
print(essay_sent_comp_cv.shape, y_cv.shape)
print(essay_sent_comp_test.shape, y_test.shape)
print("="*100)
Vectorizing Categorical features¶
- school_state : categorical data
- clean_categories : categorical data
- clean_subcategories : categorical data
- project_grade_category : categorical data
- teacher_prefix : categorical data
Vectorizing Categorical features: project grade category
In [214]:
from sklearn.feature_extraction.text import CountVectorizer
In [215]:
vectorizer = CountVectorizer(vocabulary=list(sorted_grade_dict.keys()), lowercase=False, binary=True)
vectorizer.fit(X_train['project_grade_category'].values) # fit has to happen only on train data
# we use the fitted CountVectorizer to convert the text to vector
X_train_grade_ohe = vectorizer.transform(X_train['project_grade_category'].values)
X_cv_grade_ohe = vectorizer.transform(X_cv['project_grade_category'].values)
X_test_grade_ohe = vectorizer.transform(X_test['project_grade_category'].values)
print("After vectorizations")
print(X_train_grade_ohe.shape, y_train.shape)
print(X_cv_grade_ohe.shape, y_cv.shape)
print(X_test_grade_ohe.shape, y_test.shape)
print(vectorizer.get_feature_names())
print("="*100)
Vectorizing Categorical features: teacher prefix
In [216]:
vectorizer = CountVectorizer(vocabulary=list(sorted_teacher_dict.keys()), lowercase=False, binary=True)
vectorizer.fit(X_train['teacher_prefix'].values) # fit has to happen only on train data
# we use the fitted CountVectorizer to convert the text to vector
X_train_teacher_ohe = vectorizer.transform(X_train['teacher_prefix'].values)
X_cv_teacher_ohe = vectorizer.transform(X_cv['teacher_prefix'].values)
X_test_teacher_ohe = vectorizer.transform(X_test['teacher_prefix'].values)
print("After vectorizations")
print(X_train_teacher_ohe.shape, y_train.shape)
print(X_cv_teacher_ohe.shape, y_cv.shape)
print(X_test_teacher_ohe.shape, y_test.shape)
print(vectorizer.get_feature_names())
print("="*100)
Vectorizing Categorical features: school state
In [217]:
vectorizer = CountVectorizer(vocabulary=list(sorted_state_dict.keys()), lowercase=False, binary=True)
vectorizer.fit(X_train['school_state'].values) # fit has to happen only on train data
# we use the fitted CountVectorizer to convert the text to vector
X_train_state_ohe = vectorizer.transform(X_train['school_state'].values)
X_cv_state_ohe = vectorizer.transform(X_cv['school_state'].values)
X_test_state_ohe = vectorizer.transform(X_test['school_state'].values)
print("After vectorizations")
print(X_train_state_ohe.shape, y_train.shape)
print(X_cv_state_ohe.shape, y_cv.shape)
print(X_test_state_ohe.shape, y_test.shape)
print(vectorizer.get_feature_names())
print("="*100)
Vectorizing Categorical features: clean categories
In [218]:
vectorizer = CountVectorizer(vocabulary=list(sorted_cat_dict.keys()), lowercase=False, binary=True)
vectorizer.fit(X_train['clean_categories'].values) # fit has to happen only on train data
# we use the fitted CountVectorizer to convert the text to vector
X_train_cat_ohe = vectorizer.transform(X_train['clean_categories'].values)
X_cv_cat_ohe = vectorizer.transform(X_cv['clean_categories'].values)
X_test_cat_ohe = vectorizer.transform(X_test['clean_categories'].values)
print("After vectorizations")
print(X_train_cat_ohe.shape, y_train.shape)
print(X_cv_cat_ohe.shape, y_cv.shape)
print(X_test_cat_ohe.shape, y_test.shape)
print(vectorizer.get_feature_names())
print("="*100)
Vectorizing Categorical features: clean subcategories
In [219]:
vectorizer = CountVectorizer(vocabulary=list(sorted_sub_cat_dict.keys()), lowercase=False, binary=True)
vectorizer.fit(X_train['school_state'].values) # fit has to happen only on train data
# we use the fitted CountVectorizer to convert the text to vector
X_train_sub_ohe = vectorizer.transform(X_train['clean_subcategories'].values)
X_cv_sub_ohe = vectorizer.transform(X_cv['clean_subcategories'].values)
X_test_sub_ohe = vectorizer.transform(X_test['clean_subcategories'].values)
print("After vectorizations")
print(X_train_sub_ohe.shape, y_train.shape)
print(X_cv_sub_ohe.shape, y_cv.shape)
print(X_test_sub_ohe.shape, y_test.shape)
print(vectorizer.get_feature_names())
print("="*100)
2.4 Appling Logistic Regression on different kind of featurization as mentioned in the instructions
Apply Logistic Regression on different kind of featurization as mentioned in the instructions
For Every model that you work on make sure you do the step 2 and step 3 of instrucations
Applying Logistic Regression on BOW, SET 1¶
Creating Data Matrix¶
In [220]:
# Please write all the code with proper documentation
# merge two sparse matrices: https://stackoverflow.com/a/19710648/4084039
from scipy.sparse import hstack
X_tr = hstack((X_train_essay_bow,X_train_title_bow, X_train_state_ohe, X_train_teacher_ohe, X_train_grade_ohe,X_train_cat_ohe,X_train_sub_ohe, X_train_price_norm,X_train_project_norm)).tocsr()
X_cr = hstack((X_cv_essay_bow,X_cv_title_bow, X_cv_state_ohe, X_cv_teacher_ohe, X_cv_grade_ohe,X_cv_cat_ohe,X_cv_sub_ohe, X_cv_price_norm,X_cv_project_norm)).tocsr()
X_te = hstack((X_test_essay_bow,X_test_title_bow, X_test_state_ohe, X_test_teacher_ohe, X_test_grade_ohe,X_test_cat_ohe,X_test_sub_ohe, X_test_price_norm,X_test_project_norm)).tocsr()
print("Final Data matrix")
print(X_tr.shape, y_train.shape)
print(X_cr.shape, y_cv.shape)
print(X_te.shape, y_test.shape)
print("="*100)
Hyperparameter Tuning: Simple for loop (if you are having memory limitations use this)
In [221]:
def batch_predict(clf, data):
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_data_pred = []
tr_loop = data.shape[0] - data.shape[0]%1000
# consider you X_tr shape is 49041, then your cr_loop will be 49041 - 49041%1000 = 49000
# in this for loop we will iterate unti the last 1000 multiplier
for i in range(0, tr_loop, 1000):
y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])
# we will be predicting for the last data points
y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])
return y_data_pred
In [222]:
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
from sklearn.linear_model import LogisticRegression
"""
y_true : array, shape = [n_samples] or [n_samples, n_classes]
True binary labels or binary label indicators.
y_score : array, shape = [n_samples] or [n_samples, n_classes]
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of
decisions (as returned by “decision_function” on some classifiers).
For binary y_true, y_score is supposed to be the score of the class with greater label.
"""
train_auc = []
cv_auc = []
log_alphas=[]
parameters = {'C':[0.0001,0.0005,0.001,0.005,0.01,0.05,0.1,0.5,1,2.5,5]}
for i in tqdm(parameters['C']):
neigh = LogisticRegression(C=i)
neigh.fit(X_tr, y_train)
y_train_pred = batch_predict(neigh, X_tr)
y_cv_pred = batch_predict(neigh, X_cr)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
train_auc.append(roc_auc_score(y_train,y_train_pred))
cv_auc.append(roc_auc_score(y_cv, y_cv_pred))
In [223]:
import math
for a in tqdm(parameters['C']):
b = math.log10(a)
log_alphas.append(b)
print(log_alphas)
In [224]:
plt.figure(figsize=(20,15))
plt.plot(log_alphas, train_auc, label='Train AUC')
plt.plot(log_alphas, cv_auc, label='CV AUC')
plt.scatter(log_alphas, train_auc, label='Train AUC points')
plt.scatter(log_alphas, cv_auc, label='CV AUC points')
plt.legend()
plt.xlabel("C: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
In [232]:
best_k=0.005
Train The Model
In [235]:
from sklearn.metrics import roc_curve, auc
neigh = LogisticRegression(C=best_k)
neigh.fit(X_tr, y_train)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_train_pred = batch_predict(neigh, X_tr)
y_test_pred = batch_predict(neigh, X_te)
train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)
x=[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]
In [236]:
# from sklearn.metrics import roc_curve, auc
# neigh = LogisticRegression(C=best_k)
# neigh.fit(X_tr, y_train)
# # roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# # not the predicted outputs
# y_train_pred = neigh.predict_proba(X_tr)
# y_test_pred = neigh.predict_proba(X_te)
# train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred[:,1])
# test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred[:,1])
# x=[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]
In [237]:
plt.plot(train_fpr, train_tpr, label="train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="test AUC ="+str(auc(test_fpr, test_tpr)))
plt.plot(x,x)
plt.legend()
plt.xlabel("K: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
Confusion Matrix
In [238]:
# we are writing our own function for predict, with defined thresould
# we will pick a threshold that will give the least fpr
def predict(proba, threshould, fpr, tpr):
t = threshould[np.argmax(fpr*(1-tpr))]
# (tpr*(1-fpr)) will be maximum if your fpr is very low and tpr is very high
print("the maximum value of tpr*(1-fpr)", max(tpr*(1-fpr)), "for threshold", np.round(t,3))
predictions = []
for i in proba:
if i>=t:
predictions.append(1)
else:
predictions.append(0)
return predictions
In [239]:
print("Train confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_train,predict(y_train_pred,tr_thresholds,train_fpr,train_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[239]:
In [240]:
print("Test confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_test,predict(y_test_pred,tr_thresholds,test_fpr,test_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[240]:
2.4.2 Applying Logistic Regression on TFIDF, SET 2¶
Creating Data Matrix¶
In [241]:
# Please write all the code with proper documentation
# merge two sparse matrices: https://stackoverflow.com/a/19710648/4084039
from scipy.sparse import hstack
X_tr = hstack((X_train_essay_tfidf,X_train_title_tfidf, X_train_state_ohe, X_train_teacher_ohe, X_train_grade_ohe,X_train_cat_ohe,X_train_sub_ohe, X_train_price_norm,X_train_project_norm)).tocsr()
X_cr = hstack((X_cv_essay_tfidf,X_cv_title_tfidf, X_cv_state_ohe, X_cv_teacher_ohe, X_cv_grade_ohe,X_cv_cat_ohe,X_cv_sub_ohe, X_cv_price_norm,X_cv_project_norm)).tocsr()
X_te = hstack((X_test_essay_tfidf,X_test_title_tfidf, X_test_state_ohe, X_test_teacher_ohe, X_test_grade_ohe,X_test_cat_ohe,X_test_sub_ohe, X_test_price_norm,X_test_project_norm)).tocsr()
print("Final Data matrix")
print(X_tr.shape, y_train.shape)
print(X_cr.shape, y_cv.shape)
print(X_te.shape, y_test.shape)
print("="*100)
Hyperparameter Tuning: Simple for loop (if you are having memory limitations use this)
In [242]:
def batch_predict(clf, data):
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_data_pred = []
tr_loop = data.shape[0] - data.shape[0]%1000
# consider you X_tr shape is 49041, then your cr_loop will be 49041 - 49041%1000 = 49000
# in this for loop we will iterate unti the last 1000 multiplier
for i in range(0, tr_loop, 1000):
y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])
# we will be predicting for the last data points
y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])
return y_data_pred
In [243]:
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
from sklearn.linear_model import LogisticRegression
"""
y_true : array, shape = [n_samples] or [n_samples, n_classes]
True binary labels or binary label indicators.
y_score : array, shape = [n_samples] or [n_samples, n_classes]
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of
decisions (as returned by “decision_function” on some classifiers).
For binary y_true, y_score is supposed to be the score of the class with greater label.
"""
train_auc = []
cv_auc = []
log_alphas=[]
parameters = {'C':[0.0001,0.0005,0.001,0.005,0.01,0.05,0.1,0.5,1,2.5,5,7.5,10]}
for i in tqdm(parameters['C']):
neigh = LogisticRegression(C=i)
neigh.fit(X_tr, y_train)
y_train_pred = batch_predict(neigh, X_tr)
y_cv_pred = batch_predict(neigh, X_cr)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
train_auc.append(roc_auc_score(y_train,y_train_pred))
cv_auc.append(roc_auc_score(y_cv, y_cv_pred))
In [244]:
for a in tqdm(parameters['C']):
b = math.log10(a)
log_alphas.append(b)
print(log_alphas)
In [245]:
plt.figure(figsize=(20,15))
plt.plot(log_alphas, train_auc, label='Train AUC')
plt.plot(log_alphas, cv_auc, label='CV AUC')
plt.scatter(log_alphas, train_auc, label='Train AUC points')
plt.scatter(log_alphas, cv_auc, label='CV AUC points')
plt.legend()
plt.xlabel("C: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
In [246]:
best_k=0.30
Train The Model
In [247]:
from sklearn.metrics import roc_curve, auc
neigh = LogisticRegression(C=best_k)
neigh.fit(X_tr, y_train)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_train_pred = batch_predict(neigh, X_tr)
y_test_pred = batch_predict(neigh, X_te)
train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)
x=[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]
In [248]:
plt.plot(train_fpr, train_tpr, label="train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="test AUC ="+str(auc(test_fpr, test_tpr)))
plt.plot(x,x)
plt.legend()
plt.xlabel("K: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
Confusion Matrix
In [249]:
# we are writing our own function for predict, with defined thresould
# we will pick a threshold that will give the least fpr
def predict(proba, threshould, fpr, tpr):
t = threshould[np.argmax(fpr*(1-tpr))]
# (tpr*(1-fpr)) will be maximum if your fpr is very low and tpr is very high
print("the maximum value of tpr*(1-fpr)", max(tpr*(1-fpr)), "for threshold", np.round(t,3))
predictions = []
for i in proba:
if i>=t:
predictions.append(1)
else:
predictions.append(0)
return predictions
In [250]:
print("Train confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_train,predict(y_train_pred,tr_thresholds,train_fpr,train_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[250]:
In [251]:
print("Test confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_test,predict(y_test_pred,tr_thresholds,test_fpr,test_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[251]:
2.4.3 Applying Logistic Regression on AVG W2V, SET 3¶
Creating Data Matrix¶
In [252]:
# Please write all the code with proper documentation
# merge two sparse matrices: https://stackoverflow.com/a/19710648/4084039
from scipy.sparse import hstack
X_tr = hstack((avg_w2v_essay_train,avg_w2v_title_train, X_train_state_ohe, X_train_teacher_ohe, X_train_grade_ohe,X_train_cat_ohe,X_train_sub_ohe, X_train_price_norm,X_train_project_norm)).tocsr()
X_cr = hstack((avg_w2v_essay_cv,avg_w2v_title_cv, X_cv_state_ohe, X_cv_teacher_ohe, X_cv_grade_ohe,X_cv_cat_ohe,X_cv_sub_ohe, X_cv_price_norm,X_cv_project_norm)).tocsr()
X_te = hstack((avg_w2v_essay_test,avg_w2v_title_test, X_test_state_ohe, X_test_teacher_ohe, X_test_grade_ohe,X_test_cat_ohe,X_test_sub_ohe, X_test_price_norm,X_test_project_norm)).tocsr()
print("Final Data matrix")
print(X_tr.shape, y_train.shape)
print(X_cr.shape, y_cv.shape)
print(X_te.shape, y_test.shape)
print("="*100)
Hyperparameter Tuning: Simple for loop (if you are having memory limitations use this)
In [253]:
def batch_predict(clf, data):
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_data_pred = []
tr_loop = data.shape[0] - data.shape[0]%1000
# consider you X_tr shape is 49041, then your cr_loop will be 49041 - 49041%1000 = 49000
# in this for loop we will iterate unti the last 1000 multiplier
for i in range(0, tr_loop, 1000):
y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])
# we will be predicting for the last data points
y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])
return y_data_pred
In [254]:
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
from sklearn.linear_model import LogisticRegression
"""
y_true : array, shape = [n_samples] or [n_samples, n_classes]
True binary labels or binary label indicators.
y_score : array, shape = [n_samples] or [n_samples, n_classes]
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of
decisions (as returned by “decision_function” on some classifiers).
For binary y_true, y_score is supposed to be the score of the class with greater label.
"""
train_auc = []
cv_auc = []
log_alphas=[]
parameters = {'C':[0.01,0.05,0.1,0.5,1,5,10,50,100,500,1000,5000,10000,50000]}
for i in tqdm(parameters['C']):
neigh = LogisticRegression(C=i)
neigh.fit(X_tr, y_train)
y_train_pred = batch_predict(neigh, X_tr)
y_cv_pred = batch_predict(neigh, X_cr)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
train_auc.append(roc_auc_score(y_train,y_train_pred))
cv_auc.append(roc_auc_score(y_cv, y_cv_pred))
In [257]:
for a in tqdm(parameters['C']):
b = math.log10(a)
log_alphas.append(b)
print(log_alphas)
In [258]:
plt.figure(figsize=(20,15))
plt.plot(log_alphas, train_auc, label='Train AUC')
plt.plot(log_alphas, cv_auc, label='CV AUC')
plt.scatter(log_alphas, train_auc, label='Train AUC points')
plt.scatter(log_alphas, cv_auc, label='CV AUC points')
plt.legend()
plt.xlabel("C: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
In [259]:
best_k=1
Train The Model
In [260]:
from sklearn.metrics import roc_curve, auc
neigh = LogisticRegression(C=best_k)
neigh.fit(X_tr, y_train)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_train_pred = batch_predict(neigh, X_tr)
y_test_pred = batch_predict(neigh, X_te)
train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)
In [261]:
x=[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]
plt.plot(train_fpr, train_tpr, label="train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="train AUC ="+str(auc(test_fpr, test_tpr)))
plt.plot(x,x)
plt.legend()
plt.xlabel("K: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
Confusion Matrix
In [262]:
# we are writing our own function for predict, with defined thresould
# we will pick a threshold that will give the least fpr
def predict(proba, threshould, fpr, tpr):
t = threshould[np.argmax(fpr*(1-tpr))]
# (tpr*(1-fpr)) will be maximum if your fpr is very low and tpr is very high
print("the maximum value of tpr*(1-fpr)", max(tpr*(1-fpr)), "for threshold", np.round(t,3))
predictions = []
for i in proba:
if i>=t:
predictions.append(1)
else:
predictions.append(0)
return predictions
In [263]:
print("Train confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_train,predict(y_train_pred,tr_thresholds,train_fpr,train_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[263]:
In [264]:
print("Test confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_test,predict(y_test_pred,tr_thresholds,test_fpr,test_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[264]:
2.4.4 Applying Logistic Regression on TFIDF W2V, SET 4¶
Creating Data Matrix¶
In [265]:
# Please write all the code with proper documentation
# merge two sparse matrices: https://stackoverflow.com/a/19710648/4084039
from scipy.sparse import hstack
X_tr = hstack((tfidf_w2v_train_essay,tfidf_w2v_train_title, X_train_state_ohe, X_train_teacher_ohe, X_train_grade_ohe,X_train_cat_ohe,X_train_sub_ohe, X_train_price_norm,X_train_project_norm)).tocsr()
X_cr = hstack((tfidf_w2v_cv_essay,tfidf_w2v_cv_title, X_cv_state_ohe, X_cv_teacher_ohe, X_cv_grade_ohe,X_cv_cat_ohe,X_cv_sub_ohe, X_cv_price_norm,X_cv_project_norm)).tocsr()
X_te = hstack((tfidf_w2v_test_essay,tfidf_w2v_test_title, X_test_state_ohe, X_test_teacher_ohe, X_test_grade_ohe,X_test_cat_ohe,X_test_sub_ohe, X_test_price_norm,X_test_project_norm)).tocsr()
print("Final Data matrix")
print(X_tr.shape, y_train.shape)
print(X_cr.shape, y_cv.shape)
print(X_te.shape, y_test.shape)
print("="*100)
Hyperparameter Tuning: Simple for loop (if you are having memory limitations use this)
In [266]:
def batch_predict(clf, data):
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_data_pred = []
tr_loop = data.shape[0] - data.shape[0]%1000
# consider you X_tr shape is 49041, then your cr_loop will be 49041 - 49041%1000 = 49000
# in this for loop we will iterate unti the last 1000 multiplier
for i in range(0, tr_loop, 1000):
y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])
# we will be predicting for the last data points
y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])
return y_data_pred
In [267]:
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
from sklearn.linear_model import LogisticRegression
"""
y_true : array, shape = [n_samples] or [n_samples, n_classes]
True binary labels or binary label indicators.
y_score : array, shape = [n_samples] or [n_samples, n_classes]
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of
decisions (as returned by “decision_function” on some classifiers).
For binary y_true, y_score is supposed to be the score of the class with greater label.
"""
train_auc = []
cv_auc = []
log_alphas=[]
parameters = {'C':[0.0001,0.0005,0.001,0.005,0.01,0.05,0.1,0.5,1,2.5,5,10,50,100,500]}
for i in tqdm(parameters['C']):
neigh = LogisticRegression(C=i)
neigh.fit(X_tr, y_train)
y_train_pred = batch_predict(neigh, X_tr)
y_cv_pred = batch_predict(neigh, X_cr)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
train_auc.append(roc_auc_score(y_train,y_train_pred))
cv_auc.append(roc_auc_score(y_cv, y_cv_pred))
In [268]:
for a in tqdm(parameters['C']):
b = math.log10(a)
log_alphas.append(b)
print(log_alphas)
In [269]:
plt.figure(figsize=(20,15))
plt.plot(log_alphas, train_auc, label='Train AUC')
plt.plot(log_alphas, cv_auc, label='CV AUC')
plt.scatter(log_alphas, train_auc, label='Train AUC points')
plt.scatter(log_alphas, cv_auc, label='CV AUC points')
plt.legend()
plt.xlabel("C: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
In [270]:
best_k=0.1
Train The Model
In [271]:
from sklearn.metrics import roc_curve, auc
neigh = LogisticRegression(C=best_k)
neigh.fit(X_tr, y_train)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_train_pred = batch_predict(neigh, X_tr)
y_test_pred = batch_predict(neigh, X_te)
train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)
In [272]:
x=[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]
plt.plot(train_fpr, train_tpr, label="train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="train AUC ="+str(auc(test_fpr, test_tpr)))
plt.plot(x,x)
plt.legend()
plt.xlabel("K: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
Confusion Matrix
In [273]:
# we are writing our own function for predict, with defined thresould
# we will pick a threshold that will give the least fpr
def predict(proba, threshould, fpr, tpr):
t = threshould[np.argmax(fpr*(1-tpr))]
# (tpr*(1-fpr)) will be maximum if your fpr is very low and tpr is very high
print("the maximum value of tpr*(1-fpr)", max(tpr*(1-fpr)), "for threshold", np.round(t,3))
predictions = []
for i in proba:
if i>=t:
predictions.append(1)
else:
predictions.append(0)
return predictions
In [274]:
print("Train confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_train,predict(y_train_pred,tr_thresholds,train_fpr,train_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[274]:
In [275]:
print("Test confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_test,predict(y_test_pred,tr_thresholds,test_fpr,test_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[275]:
2.5 Logistic Regression with added Features, SET 5¶
Creating Data Matrix¶
In [276]:
X_tr = hstack((X_train_state_ohe, X_train_teacher_ohe, X_train_grade_ohe,X_train_cat_ohe,X_train_sub_ohe, X_train_price_norm,X_train_project_norm,X_train_title_norm,X_train_essay_norm,essay_sent_pos_train,essay_sent_neg_train,essay_sent_neu_train,essay_sent_comp_train)).tocsr()
X_cr = hstack((X_cv_state_ohe, X_cv_teacher_ohe, X_cv_grade_ohe,X_cv_cat_ohe,X_cv_sub_ohe, X_cv_price_norm,X_cv_project_norm,X_cv_title_norm,X_cv_essay_norm,essay_sent_pos_cv,essay_sent_neg_cv,essay_sent_neu_cv,essay_sent_comp_cv)).tocsr()
X_te = hstack((X_test_state_ohe, X_test_teacher_ohe, X_test_grade_ohe,X_test_cat_ohe,X_test_sub_ohe, X_test_price_norm,X_test_project_norm,X_test_title_norm,X_test_essay_norm,essay_sent_pos_test,essay_sent_neg_test,essay_sent_neu_test,essay_sent_comp_test)).tocsr()
print("Final Data matrix")
print(X_tr.shape, y_train.shape)
print(X_cr.shape, y_cv.shape)
print(X_te.shape, y_test.shape)
print("="*100)
Hyperparameter Tuning: Simple for loop (if you are having memory limitations use this)
In [277]:
def batch_predict(clf, data):
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_data_pred = []
tr_loop = data.shape[0] - data.shape[0]%1000
# consider you X_tr shape is 49041, then your cr_loop will be 49041 - 49041%1000 = 49000
# in this for loop we will iterate unti the last 1000 multiplier
for i in range(0, tr_loop, 1000):
y_data_pred.extend(clf.predict_proba(data[i:i+1000])[:,1])
# we will be predicting for the last data points
y_data_pred.extend(clf.predict_proba(data[tr_loop:])[:,1])
return y_data_pred
In [278]:
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
from sklearn.linear_model import LogisticRegression
"""
y_true : array, shape = [n_samples] or [n_samples, n_classes]
True binary labels or binary label indicators.
y_score : array, shape = [n_samples] or [n_samples, n_classes]
Target scores, can either be probability estimates of the positive class, confidence values, or non-thresholded measure of
decisions (as returned by “decision_function” on some classifiers).
For binary y_true, y_score is supposed to be the score of the class with greater label.
"""
train_auc = []
cv_auc = []
log_alphas=[]
parameters = {'C':[0.0001,0.0005,0.001,0.005,0.01,0.05,0.1,0.5,1,2.5,5]}
for i in tqdm(parameters['C']):
neigh = LogisticRegression(C=i)
neigh.fit(X_tr, y_train)
y_train_pred = batch_predict(neigh, X_tr)
y_cv_pred = batch_predict(neigh, X_cr)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
train_auc.append(roc_auc_score(y_train,y_train_pred))
cv_auc.append(roc_auc_score(y_cv, y_cv_pred))
In [279]:
import math
for a in tqdm(parameters['C']):
b = math.log10(a)
log_alphas.append(b)
print(log_alphas)
In [280]:
plt.figure(figsize=(20,15))
plt.plot(log_alphas, train_auc, label='Train AUC')
plt.plot(log_alphas, cv_auc, label='CV AUC')
plt.scatter(log_alphas, train_auc, label='Train AUC points')
plt.scatter(log_alphas, cv_auc, label='CV AUC points')
plt.legend()
plt.xlabel("C: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
In [281]:
best_k=0.01
Train The Model
In [282]:
from sklearn.metrics import roc_curve, auc
neigh = LogisticRegression(C=best_k)
neigh.fit(X_tr, y_train)
# roc_auc_score(y_true, y_score) the 2nd parameter should be probability estimates of the positive class
# not the predicted outputs
y_train_pred = batch_predict(neigh, X_tr)
y_test_pred = batch_predict(neigh, X_te)
train_fpr, train_tpr, tr_thresholds = roc_curve(y_train, y_train_pred)
test_fpr, test_tpr, te_thresholds = roc_curve(y_test, y_test_pred)
In [283]:
x=[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]
plt.plot(train_fpr, train_tpr, label="train AUC ="+str(auc(train_fpr, train_tpr)))
plt.plot(test_fpr, test_tpr, label="train AUC ="+str(auc(test_fpr, test_tpr)))
plt.plot(x,x)
plt.legend()
plt.xlabel("K: hyperparameter")
plt.ylabel("AUC")
plt.title("ERROR PLOTS")
plt.grid()
plt.show()
Confusion Matrix
In [284]:
# we are writing our own function for predict, with defined thresould
# we will pick a threshold that will give the least fpr
def predict(proba, threshould, fpr, tpr):
t = threshould[np.argmax(fpr*(1-tpr))]
# (tpr*(1-fpr)) will be maximum if your fpr is very low and tpr is very high
print("the maximum value of tpr*(1-fpr)", max(tpr*(1-fpr)), "for threshold", np.round(t,3))
predictions = []
for i in proba:
if i>=t:
predictions.append(1)
else:
predictions.append(0)
return predictions
In [285]:
print("Train confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_train,predict(y_train_pred,tr_thresholds,train_fpr,train_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[285]:
In [286]:
print("Test confusion matrix")
conf_matr_df_train_2=pd.DataFrame(confusion_matrix(y_test,predict(y_test_pred,tr_thresholds,test_fpr,test_fpr)),range(2),range(2))
sns.set(font_scale=1)#for label size
sns.heatmap(conf_matr_df_train_2,annot=True,annot_kws={"size":30},fmt='g')
Out[286]:
3. Conclusion
In [288]:
# Please compare all your mod# http://zetcode.com/python/prettytable/
from prettytable import PrettyTable
#If you get a ModuleNotFoundError error , install prettytable using: pip3 install prettytable
x=PrettyTable()
x.field_names=["Vectorizer","Hyper Parameter","AUC"]
x.add_row(["BOW",0.01,0.66])
x.add_row(["TFIDF",0.32,0.65])
x.add_row(["AVG W2V",1,0.69])
x.add_row(["TFIDF W2V",0.1,0.68])
x.add_row(["WITHOUT TEXT",0.01,0.57])
print(x)
INFERENCE:¶
Here we can be observed that "Essays" and "Project Titles" play a major role in predicting the outcome of the project. So,we should not neglect them as the top four models containing them proved to have a better AUC score.
In [ ]: